From: AAAI-02 Proceedings. Copyright © 2002, AAAI (www.aaai.org). All rights reserved.
Invited Talks and Panels
2002 AAAI Presidential Address:
AI and the Impending Revolution in Brain Science
Tom M. Mitchell, Carnegie Mellon University
AAAI-02 Keynote Address:
Probabilistic AI and Information Retrieval
Michael I. Jordan, University of California, Berkeley
Much progress has been made in recent years in the area of
information retrieval, in particular as embodied in Internet
search engine technology. Much progress has also been
made in probabilistic, graph-theoretic AI. What are the possibilities for bringing these two lines of research together —
for viewing large-scale information retrieval as a core enabling technology for AI systems, and for asking IR systems
to exhibit true inferential capabilities? Jordan will discuss
research aimed at bridging the AI/IR gap.
AAAI-02/IAAI-02 Joint Invited Talk:
Human Level "Strong" AI: The Prospects and Implications
Raymond Kurzweil, KurzweilAI.net (Kurzweil Accelerating Intelligence Network)
Three-dimensional molecular computing will provide the
hardware for human-level "strong" AI well before 2030.
The more important software insights will be gained in part
from the reverse-engineering of the human brain, a process
well under way. Once nonbiological intelligence matches
the range and subtlety of human intelligence, it will neces-
sarily soar past it because of the continuing acceleration of
information-based technologies, as well as the ability of
machines to instantly share their knowledge. The implication will be an intimate merger between the technologycreating species and the evolutionary process it spawned.
xxv
AAAI-02 Invited Talk:
Perspectives on Artificial Intelligence Planning
Hector Geffner, ICREA - Universitat Pompeu Fabra (Barcelona)
Planning has always been a key area in artificial intelligence. In its general form, planning is concerned with the
automatic synthesis of action strategies (plans) from a description of actions, sensors, and goals. Planning thus contrasts with two other approaches to intelligent behavior: the
programming approach, where action strategies are defined
by hand, and the learning approach, where action strategies
are inferred from experience.
Different assumptions about the nature of actions, sensors, and costs lead to various forms of planning: (1) Planning with complete information and deterministic actions,
(2) planning with non-deterministic actions and sensing, and
(3) planning with temporal and concurrent actions, etc. Most
work so far has been devoted to "classical" planning (1.
above), where significant changes have taken place in the
last few years. On the methodological side, the area has become more empirical with experimental evaluation being
routine; on the technical side, approaches based on heuristic
or constrained-based search have taken over blind-search
approaches.
In this talk, Geffner will provide a coherent picture of
planning in AI while trying to convey some of the current
excitement in the field. He'll make emphasis on the mathematical models that underlie various forms of planning, and
the ideas that have been found most useful computationally.
AAAI-02 Invited Talk:
Dimension Reduction that Preserves Information and Neural Coding
Naftali Tishby, The Hebrew University
Many cognitive functions, such as prediction, feature extraction, noise filtering, and learning, can be viewed as special cases of one principle: compression while preserving
information. This information theoretic principle was turned
into a computational paradigm: the information bottleneck
method. This variational method yielded several novel
learning and data analysis algorithms, with many applications to information retrieval as well as to analysis of neural
coding in several neurobiological systems, that were carried
in Tishby's lab. In this talk Tishby will focus on a new approach to data dimensionality reduction that stems from this
principle. Here he searches for low dimensional (nonlinear)
reduction of co-occurrence (or contingency) tables that preserve the (mutual) information in the table. He gives a new
alternate-projection algorithm for achieving such a reduction and show its convergence to an optimal set of information preserving features. This approach is particularly useful
when the data is not naturally quantized but rather represented by low dimension continuous features. Such a reduction may have interesting biological implications. (Based on
joint work with Amir Globerson and Noam Slonim.)
IAAI-02 Invited Talk:
Robot-Assisted Urban Search and Rescue at the WTC: Where's the AI?
Robin R. Murphy, University of South Florida
On September 11, 2001, the Center for Robot-Assisted
Search and Rescue responded within six hours to the WTC
disaster; this is the first known use of robots for USAR. The
University of South Florida was one of the four robot teams,
and only academic institution. The USF team participated
on-site in the search efforts from September 12 through 22,
collecting and archiving data on the use of robots.
This talk will provide an overview of the use of robots for
USAR as well as discuss what AI techniques were available,
what was actually used, and why. It will also summarize the
key lessons learned from the robotics efforts at the WTC.
xxvi
The lessons learned cover the areas of platforms and mobility, sensors and sensing strategies, control, and human-robot
interactions. Possibly the most pervasive lesson learned is
that robots for USAR must be considered from an "information technology" perspective, where platforms, sensors,
control schemes, networks, and interfaces must all be
coevolved to ensure the information extracted by the robots
is truly usable by the rescue community.
Extensive video footage of the site and "robot's eye"
views will be shown.
IAAI-02 Invited Panel:
Pioneering AI Businesses I: A 20-Year Review
Panel Leader: Neil Jacobstein, Teknowledge Corporation
Several AI-based businesses started in the early 1980s. They
underwent a classic boom and bust cycle. Hype exceeded
expectations, and some investors and technologists lost patience. However, history shows that in cases of disruptive
technological innovation, forecasts are usually too optimistic in the short run, and too conservative in the long run. Is
that the case with AI businesses? This panel of AI entrepre-
neurs will review the technology base and history of pioneering AI businesses, extract lessons learned, and identify
future opportunities. Companies discussed will include IntelliCorp, Teknowledge, Inference, Syntelligence, Carnegie
Group, Cycorp, and others. An interactive question and answer session with panel members will follow brief presentations from each panelist.
IAAI-02 Invited Panel:
Pioneering AI Businesses II: Recent Startups
Panel Leader: Craig Knoblock, University of Southern California and Fetch Technologies
This panel will focus on the process of starting an AI company. The barriers to starting a new company include applying the technology to address a specific market need and
creating and running a successful business. The speakers on
this panel are AI researchers that have recently started their
own companies. Some of the issues to be discussed by the
panelists include how to go from a technology to a business,
how to get funding for a company, and what are some of the
pitfalls to watch out for. An interactive question and answer
session with panel members will follow brief presentations
from each panelist.
IAAI-02 Invited Panel (Collocated with AI Festival):
PI Entrepreneurs Forum
This forum will provide an open and informal setting for AI
pioneers, technologists, entrepreneurs, venture capitalists,
legal, and intellectual property experts to network and discuss issues in starting and running AI-based companies.
xxvii